fmin map

[Kodiweera]

Hi Gary,

What is fmin map? why it has such high numbers?

Thanks much

Chandana

[Hui Zhang]

Hi Chandana,

It is the (local) minimum of the fitting objective function.  The actual values depend on the specifics of each dataset.

Gary

--
You received this message because you are subscribed to the Google Groups "NODDI" group.
To unsubscribe from this group and stop receiving emails from it, send an email to noddi+un...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

[Kodiweera]

Thank you Gary.

-Chandana

[Francesco Grussu]

Hello Gary,

I am in the process of studying NODDI quality of fit within two ROIs. I want to tell in which one of the two the model is being fitted best, and I thought to study fmin for such a purpose.

I seem to understand that I should choose the ROI where fmin is overall smaller, since it is the (local) minimum of the objective function.
However, in my case I get fmin which is negative and ranging from roughly -4E+6 to -0.5E+6. This makes me think of a kind of Log-Likelihood type of measure, which should therefore be maximised, as opposed to what I seem to understand from this thread. 

I am using the latest version of the NODDI toolbox (version 0.9).

Could you possibly clarify about this?

Many thanks indeed for your help,
best wishes

Francesco

[Hui Zhang]

Hi Francesco,

The fmin is the opposite of the Rician Log-likelihood of the model fitting.  The model fitting is indeed achieved with the standard maximum-likelihood approach.  But because we are using a minimiser, we have to use an objective function that is the negated Rician Log-likelihood.  So for the purpose of assessing the quality of fitting, you should use the value from negating the reported fmin.

I hope this make sense.

Gary

For more options, visit https://groups.google.com/d/optout.

[Francesco Grussu]

It does make sense, thanks indeed.
Therefore, it should also hold that I can calculate the Bayesian Information Criterion (BIC) simply as

BIC = - 2*fmin + P*log(M)

being P the number of parameters being fitted by the toolbox and being M the number of measurements?

Regards,
Francesco

[Francesco Grussu]

Sorry, I meant

BIC = + 2*fmin + P*log(M)

since  fmin = - log(L) for what you said.

[Hui Zhang]

That is correct.

Gary

[Francesco Grussu]

This is great and very helpful.

Many thanks,
regards

Francesco

That is correct.

To unsubscribe from this group and stop receiving emails from it, send an email to noddi+unsubscribe@googlegroups.com.

[Hui Zhang]

Sorry.  I need to correct myself here.  fmin is in fact equal to log(L).  So BIC = -2 fmin + P*log(M).

Thanks to Francesco for alerting me to this.

Gary

[Clio Gonzalezzacarias]

Hi Dear NODDI team,

I have a question, I am new running NODDI. In fact, I just did the small tutorial in here: http://mig.cs.ucl.ac.uk/index.php?n=Tutorial.NODDImatlab 

and I used the demo data provided and ran NODDI toolbox version 1.0.5.

I believe the maps I am getting are exactly the same as in the "NODDI_example_dataset/output"  except for the fmin map.

I am sending here screenshots of the fmin map intensity  provided and what I got.

The difference seems to be that for the result provided minimum intensity is 0 and max intensity is 4677175

for my result minimum intensity is  -4677175 and max intensity is 0.

Looks like a change in sign?  and I am wondering I this could be a problem? or something that I did wrong? 

Also, I am using this trend as it appears to be the only one about fmin maps.

Thanks in advance for your help.

Clio Gonzalez

[Hui Zhang]

Thanks for trying there NODDI toolbox, Clio, and for flagging this.  You got the right results. The discrepancy is my mistake.  I forgot to update the example outputs after making some adjustments to the code.

Gary

To view this discussion on the web visit https://groups.google.com/d/msgid/noddi/e06d1da5-5c1f-4326-b483-b15901ab2acfn%40googlegroups.com.

[Clio Gonzalezzacarias]

Thanks Gary for your help. 

All the best,

Clio